低倍率镜检图像无标记红白细胞识别方法研究

针对低倍率镜检图像无标记红白细胞对比度低、边缘模糊、结构不清晰、内部纹理特征不明显等特点，提出一种改进的低维特征向量识别算法。该算法通过彩色图像空间分离、逻辑或运算和形态学处理完成初步分割，针对粘连细胞，采用基于迭代腐蚀的标记分水岭方法再次分割，通过多种方法互补完成红白细胞的分割。根据红白细胞在形态、快速傅里叶变化(fast Fourier transform,FFT)以及Canny边缘检测图像的差别，提取周长、面积、FFT后的圆形度、连通域数、像素和和闭合比值6个相关特征组成特征向量用于训练支持向量机分类器。实验结果表明，在低倍率镜检图像无标记红白细胞准确分割前提下，基于6个相关特征的低维特征向量识别可以显著提高识别率，而且识别效果相对稳定，不易受红白细胞异型情况影响。

Study on recognition method of label-free red and white cell using low-rate microscopic image

According to the characteristics such as low contrast, fuzzy edge, unclear structure, and non-obvious internal texture features of unlabeled red and white cells in the low-rate microscopic image, an improved low dimensional feature vector recognition algorithm is proposed in this paper. Firstly, this algorithm adopts color image space separation, logic or operation and morphological processing to complete the initial segmentation, and the adherent cells are segmented again by the label watershed method based on iterative corrosion, and the segmentation of red and white cells was completed by various methods. Then according to the red and white cells in morphology, Fast Fourier transform (FFT) and Canny image edge detection difference, obtained six related features of the perimeter, area, circularity after the FFT, the number of connected domains, the pixels and the closed ratio to form a feature vector for training support vector machine classifier. The experimental results indicate that under the premise of accurate segmentation of unlabeled red and white cells in the low-rate microscopic images, the low dimensional feature vector recognition algorithm based on six correlation features can significantly improve the recognition rate, and the recognition effect is relatively stable, which is not easily affected by the abnormal red and white cells.